Background

Immunotherapy has emerged as a groundbreaking approach in cancer treatment, notably exemplified in the context of melanoma. Checkpoint blockade immunotherapy (CBI) disrupts inhibitory signals exploited by tumors to evade immune detection, harnessing the immune system to eliminate cancer cells. This results in enduring responses and extended survival for a significant fraction of melanoma patients.1 As CBI gains recognition as a primary therapy for unresectable or metastatic melanoma, understanding the intricate landscape of tumor-infiltrating immune cells becomes imperative for tailoring effective treatment strategies.

Transcriptomic studies have revealed several subsets of tumor-infiltrating CD8 T cells (CD8 TIL), most of which express PD1, categorizing them as exhausted.2 Despite the expression of exhaustion markers, these PD1+ CD8 TILs remain functional and efficient in controlling cancer cells and chronic infections.3 A distinctive differentiation pathway has been observed, going from progenitor cells to cells with enhanced cytotoxicity, ultimately leading to dysfunctional cells unable to control tumor growth.4 5 Notably, the transcription factor T cell factor 7 (TCF7 encodes for TCF1) is expressed at progenitor stages, and is the population giving rise to CD8 T cells with enhanced cytotoxic activity, which eventually become terminally exhausted after accumulating expression of several checkpoint inhibitors and displaying evidence of altered metabolism with mitochondrial dysfunction.4–7

Tissue-resident memory (TRM) CD8 T cells are normal inhabitants of peripheral tissues, including tumor sites. Numerous studies highlight their strategic location, enabling rapid and targeted responses to pathogens and cancer cells, contributing to site-specific immunosurveillance. TRM CD8 T cells exhibit superior cytotoxic and cytokine expression activity compared with non-TRM CD8 T cells,8–10 making them a key immune cell in cancer control, as it has been demonstrated in both experimental murine melanoma models and clinical studies.11–13 TRM T cells residing in tumors often express a variety of checkpoint inhibitors while still exhibiting evidence of effector activity.8 9 14 15 Since immunotherapy relies on the expression of checkpoint inhibitors, TRM CD8 T cells emerge as a key population for fostering and recovering their effector activity, making them crucial for the success of immunotherapeutic interventions. Indeed, recent investigations underscore the importance of TRM T cells in achieving CBI-positive responses.12 16

It is currently uncertain whether tumor-infiltrating TRM T cells can be separated into progenitor, effector and terminally exhausted subsets, as documented for memory T cells challenged with chronic-infecting viruses.5 6 17–19 This categorization is important, as the progenitor population seems critical to respond to CBI, by expanding and forming effector cells. Human transcriptomic studies and murine cancer and infection models support that this expanding progenitor population is characterized by expression of TCF7.5 6 20–23 However, the existence of TCF1+TRM CD8 T cells remains uncertain, as two transcription factors master of TRM formation, HOBIT and BLIMP1, have been shown to correlate with loss of TCF7 expression.24 25 Similarly, TCF7 seems to negatively regulate the expression of residency marker CD103, altogether suggesting that CBI-responding TCF1+CD8+ T cells are distinct from CBI-responding canonical TRM CD8 T cells.26 27 On the contrary, other studies support the coexpression of the TRM program and TCF1/TCF7, and we have recently identified a TCF1+CD103+CD8+ population in the tumor stroma of murine melanoma.28–32 Overall, these data pose an intriguing avenue for further exploration, particularly when aiming to thoroughly dissect the populations responding to CBI to tailor strategies to non-responding patients.

In the intricate orchestration of immune responses, the role of antigen-presenting cells is paramount. Classic dendritic cell type 1 (cDC1), with their ability to adeptly cross-present antigens to CD8+T cells, occupy a central position in triggering effective anti-tumor response.33 In mouse knockout of BATF3 that cannot form cDC1, anti-tumor CD8 cytotoxic responses are abated.34 Anti-PD1 treatment plus FLT3L stimulation expands cDC1 and protects mice against melanoma rechallenge.35 We and others have shown that the abundance of tumor-infiltrating cDC1 correlates with favorable clinical outcomes in melanoma patients.36–38 Overall, these studies support the importance of cDC1 in cancer control. However, the participation of cDC1 in cancer control on CBI treatment has not been demonstrated.

Leveraging multiparametric proteomic approaches, this study unveils the significance of TRM CD8 T cells and cDC1, demonstrating their correlation with disease prognosis and with responding patients to immunotherapy. Notably, within the TRM CD8 T cells, we identified TCF1+ and TCF1− subsets, both exerting protective effects. Our study also reveals a transformative shift in the TRM subsets as patients progress to metastasis, particularly TCF1− TRM acquire heightened expression of checkpoint inhibitory receptors and diminish expression of effector and tissue residency markers. Essentially, this study provides a thorough examination of the complex dynamics within cancer-protective immune populations. Additionally, it underscores specific immune cells, offering valuable insights for customizing precision immunotherapies to improve treatment effectiveness.

MethodsMelanoma patients and control samples

This is a multicentric study involving the formation of two patient cohorts derived from the Pathology Department and/or Oncology Services of several oncology centers: Hospital de Oncología Centro Médico Nacional Siglo XXI (CMN SXXI IMSS), Hospital de Alta Especialidad 25 (IMSS), Centro Médico Nacional Manuel Ávila Camacho (IMSS), Hospital General de Zona 1A (IMSS), Centro Médico ABC, Hospital de Alta Especialidad 25 (IMSS) and Unidad Médica de Atención Ambulatoria (UMAA 199). The first cohort comprised 58 archived paraffin blocks from melanoma resection procedures while the second cohort included 24 paraffin blocks obtained before the initiation of immunotherapy. Additionally, five fresh melanoma samples were obtained from surgical resections at Hospital de Oncología CMN SXXI IMSS. Finally, eight control skin samples were derived from archived paraffin blocks, and five were obtained from fresh biopsies. All control skin samples were from patients who underwent surgeries unrelated to autoimmune/inflammatory disorders or cancer. See online supplemental tables 1 and 2 for the patient’s clinical description.

Cultures of biopsies

Skin was cultured in supplemented RPMI 1640 medium (see online supplemental table 3 for a description of all used reagents, material, equipment and software). After a 7-day incubation period, cells that migrated from the skin were harvested, washed, and cryopreserved for future assays. TILs were obtained using a previously described method.37 Briefly, the tumor was finely minced and incubated with 400 (U/mL) Collagenase IV and (5 mg/mL) DNAse for 1 hour at 37°C. Next, enzymatic digestion was stopped by adding 0.5 µM EDTA, and digested tissues were filtered through a 70 µm strainer. After that, the cell suspension was treated with 0.2 mg/mL DNase. The lymphocyte interface of the centrifuged Percoll 40/90 solution was recovered, washed, and cryopreserved.

Immunofluorescence staining

Three µm tissue sections were affixed to electrocharged slides. Slides underwent paraffin removal, tissue rehydration, antigen retrieval, permeabilization and were incubated overnight with the respective primary antibodies. Subsequently, secondary antibodies were incubated for 2 hours. Nuclei were stained with Hoechst for 10 min and mounted with Vectashield. For multiplexed immunofluorescence (mIF), a series of staining cycles were performed on the same slide. Unconjugated primary antibodies and labeled secondary antibodies were used in the first cycle, followed by fluorochrome-conjugated antibodies in subsequent cycles to reduce cross reactivity. Tissues were mounted with PBS-Glycerol 10%, and confocal microscopy was employed for image acquisition. Fluorochromes were bleached using H2O2 (4.5%) and NaOH (24 mM) for 60 min in the presence of white light. Bleaching efficiency was confirmed under the fluorescence microscope. online supplemental file 2 shows the antibody combination of each staining with example images of the staining and bleaching steps.

Confocal microscopy

Micrographs were captured using a Nikon Ti Eclipse inverted confocal microscope with NIS Elements V.4.50 software, 20× (dry, NA 0.75) objective lens, and Nyquist’s magnification (3.4×). Three areas with high immune infiltrate, identified through H&E staining, were selected from each patient to quantify density and percentage of all phenotypes of interest. Whole slide scanning was performed with the Acquire function of the NISV.4.50 software, selecting the Scan Large Image option. Automated image overlapping and further analysis were conducted using FIJI ImageJ Software.

Fluorescence images analysis

IF and mIF analyses were conducted using a previously described in-house machine learning method.36 Python scripts were employed for further analysis of the obtained results. To quantify the percentage of positive cells, the total number of cells positive for a specific phenotype was divided by the total cell count in the field, multiplied by 100. Density was determined by dividing the number of cells of interest by the total area of the field. Results presented in the plots represent the median value from three distinct areas per sample or patient. For single-color overlap analysis, we show the areas of marker colocalization eliminating non-colocalized pixels to facilitate the visualization of cells and areas with marker coexpression. This was achieved using the Image Calculator function in FIJI ImageJ Software.

Histocytometry

The machine learning methodology generated individual .csv files for each field of every patient, encompassing measurements of individual cells. These files were concatenated through Python scripts and subsequently analyzed using BD FlowJo V.10.9 software. online supplemental file 2 shows all the gating strategies and establishment of expression cut-off points of all the evaluated markers.

Spatial analysis

To evaluate the spatial distribution of cell phenotypes within tumor and stromal regions, we created individual regional masks for each image by tracking the tumorous area with MART-1+ staining and concatenating masks to the Machine Learning mIF images analysis. Using histocytometry, we generated XY coordinate maps and imported each phenotype of interest along with their coordinates in the image. Distances between cytotoxic cells and apoptotic melanoma cells were measured by generating .csv files and importing them into CytoMAP software V.1.4.21. The calculate distance function was employed to estimate distances between Casp3-p+MART-1+ cells and TCF1+ or TCF1− TRMs. To standardize the data and compare distances from different datasets, we used the Bar and Violin plot extension, standardizing the data by subtracting the means and dividing them by the SD, as detailed in the original study.39

Flow cytometry staining

Cryopreserved cells underwent thawing and then were stained for spectral flow cytometry. Every single positive control was used to spectral unmix. Stopping gate: 10 000 CD3+CD45RO+CD8+ events. Approximately 100 000 total events were acquired of T cells in a Cytek Aurora spectral flow cytometer (3L, 16V, 14B, 8R). Acquired files were analyzed using Flowjo V.10.8 software. See online supplemental file 2 for the gating strategy. As we did not use protein transport inhibitors, we excluded cytokines data. For the unsupervised clustering, singlets, live, CD45RO, CD3, CD8, CD69, and CD103 positive events were down-sampled at 1000 events per sample and concatenated (4000 total events) as Flow Cytometry Standard (FCS) files and UMAP clusterization was performed with all compensated parameters except for, viability, CD45RO, CD3, CD8, CD4, CD69, and CD103. We used Euclidean approximation with 15 nearest neighbors, minimum distance of 0.5, and 2 number of components. An unsupervised clustering map was performed with Flowsom using the same compensated parameters. Heat maps and representative clusters were obtained with the Cluster Explorer plugin. Trajectory analyses were performed with the PHATE plugin with 15 nearest neighbors, decay of 15, and gamma 1 and 2 components.

PCA-based clustering

Principal component analysis (PCA) was performed using 12 variables measured in patients by IF. Data processing and normalization were conducted by using R V.4.2.2. Visualization was generated using the factoextra library.

Creation of TRM and cDC1 signatures

Transcriptomic signatures for TRM CD8 T cells and cDC1s were identified through a thorough search of RNA-seq or scRNA-seq studies in the NCBI PubMed database. After manual curation, eight studies proposing TRM CD8 T cell signatures and seven studies proposing cDC1 signatures were selected.40–47 Integration of these signatures revealed 36 genes shared by two or more TRM CD8 studies and 23 genes shared by cDC1 studies. Additionally, 11 TRM CD8 T cell-associated genes and 3 cDC1-associated genes were incorporated into the signature based on recent literature.40 48–59

Subsequently, we conducted a correlative expression analysis of the 47-gene (TRM CD8) and 26-gene (cDC1) signatures using The Cancer Genome Atlas (TCGA) melanoma database. We processed the RNA-seq count data for normalization, and the expression of the 47 and 26 genes of interest was converted into logCPM. The GSVA V.1.46.0 R package was employed to conduct an enrichment analysis across all TCGA melanoma samples, identifying high and low enrichment groups.60 Enriched patients were subjected to gene expression Pearson correlation analysis using the WGCNA V.1.72-1 R package,61 revealing two clusters of genes with positive and negative correlation. From these clusters, 41 TRM CD8-associated genes and 22 cDC1-associated genes showing significantly positive correlations were selected to compose comprehensive signatures. Spearman’s correlation analysis was conducted to investigate the relationship between samples enriched with TRM CD8 T cells and cDC1s.

Expression enrichment and survival association

Survival analysis and creation of Kaplan-Meier curves were performed using the survminer V.0.4.9 and survival V.3.4-0 R packages. These analyses were based on categorizing the TCGA melanoma samples into high versus low enrichment of TRM, cDC1, CD8a, CD103, and CD8a/CD103 as determined from the previous GSVA V.1.46.0 R analysis. Robustness of the results was confirmed through Cox regression analysis, which involved calculating a p value and its corresponding HR. This statistical approach was employed to compare the outcomes between the groups with high and low enrichment.

Statistical analysis

We evaluated the Gaussian distribution of all data using the D’Agostino test. For experiments involving three groups, analysis of variance with post hoc multiparametric comparison was chosen. Unpaired Student’s t-test was applied for two groups with independent quantitative variables. All statistical analyses were performed using Prisma software (GraphPad V.9). Results are presented as the mean±SEM. Since spatial distance measurements did not conform to a Gaussian distribution according to the D’Agostino test, a Mann-Whitney test was employed for this data. Statistical significance: *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001.

ResultsStudy design

We conducted a comprehensive analysis using two distinct patient cohorts (figure 1A). Cohort A included 58 paraffin-embedded tumor resection specimens from individuals diagnosed with melanoma. Among these, 24 patients were classified as disease-free (DF), while 34 patients had metastatic melanoma (M) according to the American Joint Committee on Cancer TNM system for melanoma staging.62 All samples were assessed at the time of diagnosis, and patients were subsequently monitored for a period of 2 years. In this context, “DF” denotes patients who remained in remission throughout the follow-up period, while the “metastatic” classification was assigned either at the time of diagnosis (n=21) or upon disease progression (n=3) during the follow-up period. All metastatic patients were combined due to observed similar behavior between both latter groups. Cohort B comprised tissue resection specimens from 24 patients collected before CBI treatment initiation. These patients were later classified as either responders (R, n=13) or non-responders (NR, n=11) based on RECIST response criteria, evaluated 6 months after immunotherapy commencement. Additionally, skin samples from eight individuals without neoplastic conditions served as our control group. All tissue samples underwent IF staining. In cohort A, our analysis aimed to primarily characterize CD8 T cell and cDC1 phenotypes associated with disease control and improved prognosis. Cohort B aimed to elucidate the importance of these populations for immunotherapy response. To ensure the reliability of our findings, we analyzed five additional fresh melanoma tissues and five fresh control skins, using spectral flow cytometry, to confirm the IF phenotypes observed in situ and to more in-depth characterize protective populations.

Figure 1

CD103+CD8 T cells associate with disease control. (A) Diagram of study design. We included two different cohorts to examine the immune infiltrate. In cohort A, we obtained the resection products at diagnosis. Cohort A was classified into disease-free (DF) and metastatic (M) according to their clinical evolution at 2 years. In cohort B, the resection products reflect the state before treatment with CBI, and the patients were classified as responders (R) or non-responders (NR) after 6 months of follow-up. Additionally, we included five fresh tissue resections to evaluate data by spectral flow cytometry. (B–C) Immunofluorescence staining for tumor cells (MART1+, sepia), CD8 (red), CD103 (yellow), and nuclei (blue). (B) shows representative complete fields of control skin (CS), DF, and M. (C) Dotted white squares show the area selected for digital zoom in the right column. Orange squares point out examples of CD103+CD8+ cells. (D) Percentage of CD8 T cells. (E) Tissue density of CD8 per 1×105 µm2. (F) Percentage of CD103+CD8+ T cells. (G) Tissue density of CD8+CD103+ per 1×105 µm2. Data from DF (n=24) (light blue acral patients, dark blue non-acral cutaneous melanoma) and M (n=34) (light red acral patients, dark red non-acral cutaneous melanoma) patients. Unpaired Student’s t-test, *p<0.05, **p<0.01, and ****p<0.0001.

Tumor-infiltrating CD103+ CD8 T cells are enriched in patients controlling melanoma

TILs are generally associated with improved prognoses in melanoma and other cancers. However, some patients, despite having abundant TILs, struggle to control the disease and develop metastases. This underscores the need to thoroughly characterize specific cell populations and phenotypes associated with disease control. We first focused on cytotoxic CD8 T cells, which are recognized as the primary effectors in antitumoral immune responses, and CD103 expression, a marker of tissue residency, also associated with better clinical outcomes in various cancer types.63

Although the level of cellular infiltration in DF and M melanoma samples was not strikingly different (as indicated by the abundance of nuclei in figure 1B), noticeable differences were observed in the abundance and distribution of CD8 T cells. Specifically, clusters of CD8 T cells were prominently visible in DF patients, whereas M patients exhibited fewer and more dispersed T cells throughout the samples (figure 1B,C). The analysis of three highly infiltrated regions of each sample revealed that CD103−CD8+ and CD103+CD8+ populations were significantly enriched in melanoma samples and DF patients compared with control skin (online supplemental file 2) and M patients (figure 1D–G), respectively. However, the latter appeared to better distinguish between melanoma controllers and progressors. These observations were more pronounced when the data were scored as cell densities rather than percentages. These findings suggest that, while the abundance of CD8+ T cells is crucial, the CD103+CD8+ compartment seems particularly important in effectively controlling melanoma progression.

Both TCF1+ and TCF1− CD103+ CD8 T cell subsets are protective

We previously highlighted the importance of TCF1+CD103+CD8+ cells in controlling melanoma growth in a murine model.30 Transcriptomic studies also support the existence of these cells.17 18 28 31 We assessed the abundance of these cells in tumor infiltrates from DF and M patients (figure 2A, online supplemental file 2). DF patients exhibited a significant enrichment of TCF1 CD103 double positive and CD103 single positive CD8 T cells (figure 2B, online supplemental file 2), while M patients showed enrichment in the CD103 negative subsets (online supplemental file 2). The percentage of the TCF1+CD103+ CD8 T cell subpopulation relative to the total tumor-infiltrating cells yielded similar results (figure 2C). Additionally, we found an enrichment of the TCF1−CD103+CD8+ subpopulation in DF patients (figure 2B,D). Evaluating the prognostic potential of both TCF1+ and TCF1− CD103+ CD8 subsets in relation to tumor staging revealed decreased percentages in advanced melanoma stages, emphasizing their significance in stages with favorable prognosis (see figure 2E,F).

Figure 2

CD103+ CD8 T cells are bona fide TRM cells and both TCF1− and TCF1+ subsets are protective. (A) Representative micrographs of immunofluorescence (IF) staining for TCF1 (green), CD103 (yellow), CD8 (red), and nuclei (blue). Dotted white squares show the area selected for digital zoom in the right column. White squares point out to an example of TCF1+CD103+ CD8 T cells, and yellow squares point out to TCF1− CD103+CD8 T cell. (B) Pie charts depicting the distribution of TCF1 CD103 CD8 phenotypes related to the total CD8 fraction. (C–F) Percentages of TCF1+ and TCF1− CD103+CD8 T cell populations. (C, D) Disease-free (DF) patients are in blue and metastatic (M) patients are in red. (E, F) Stage I and II (blue) and stage III and IV (red). (G) PCA plot, each blue dot represents a DF patient (n=24), and red triangles represent M patients (n=34). The resulting grouping was obtained by PCA of 12 immune cell population measurements. As a result, each measurement is denoted by black vectors with the corresponding match number from vectors: 1 (%CD8+), 2 (CD8+ tissue density), 3 (%CD103+CD8+), 4 (%TCF1+CD103+CD8+), 5 (%TCF7− CD103+CD8+), 6 (%TCF1+CD103− CD8+), 7 (%TCF1+CD103+ CD8+ of total CD8+ cells), 8 (%TCF1− CD103+CD8+ of total CD8+ cells), 9 (%TCF1+CD103− CD8+ of total CD8+ cells), 10 (%TCF1− CD103− CD8+ of total CD8+ cells), 11 (%cDC1 cells with respect to the total infiltrate) and 12 (%CD11c with respect to the total infiltrate). (C–G) Acral melanoma is presented in light blue (DF plots) or light red (M plots), non-acral cutaneous melanoma is presented in dark blue (DF plots) and dark red (M plots). (H) Multiplex immunofluorescence staining for CD8 (red), TCF1 (green), CD103 (yellow), CD45RO (cyan), CD69 (magenta), HOBIT (gray), RUNX3 (orange), and nuclei (blue). Squares point to CD103+CD69+ HOBIT+ RUNX3+CD8+ CD45RO+TCF1+ cells (white) and TCF1− cells (yellow). (I) Quantification of CD103+ cells on HOBIT+RUNX3+ CD45RO+CD8+ fraction, in both TCF1+ and TCF1− subsets. Unpaired Student’s t-test, **p<0.01, ***p<0.001 and ****p<0.0001.

Enrichment of cDC1 also characterizes melanoma controllers

Due to their crucial role in fostering the presence and activation of T cells, we next investigated cDC1 infiltration and association with CD103+ CD8 T cells. Our findings in this independent and larger cohort corroborated previous observations of the positive correlation between cDC1 and disease control (see online supplemental file 2). To reinforce our findings identifying CD103+ CD8 T cells and cDC1 associated with disease control, we compiled a comprehensive database based on our immune assessment of cohort A. This database comprised 12 variables, encompassing percentages and phenotypic compositions of immune cell subsets, along with the differentiation between DF and M outcomes. Hierarchical heatmap visualization revealed two distinct clusters, with the DF-associated cluster including the three protective populations, and the M-associated cluster composed of CD103-negative CD8 T cell subsets (see online supplemental file 2). Employing a PCA-based clustering approach, we further confirmed that CD103+CD8+ subsets and cDC1 subsets clustered with DF patients, while CD103− CD8+ subsets clustered with M patients (figure 2G). These findings support the association of both TCF1+ and TCF1− CD103+ CD8 T cells and cDC1 with melanoma control. Notably, figures 1D–G, and 2C–G, and online supplemental file 2 highlight acral melanomas, revealing an equal distribution between DF and M patients. These findings support the notion that both TCF1+ and TCF1− CD103+ CD8 T cells, along with cDC1 cells, also play protective roles in this melanoma subtype.

CD103+ CD8 T cells are bona fide TRM T cells

Recent studies have emphasized the role of TRM CD8 T cells as a key immune population associated with cancer control across various solid tumors. CD103 acts as a ligand for E-cadherin, anchoring T cells to epithelial tissue, making CD103 expression indicative of tissue residence. We aimed to thoroughly characterize the phenotype of CD103+ CD8 T cells to unequivocally establish their identity as TRM cells and confirm the evidence that both TCF1+ and TCF1− CD103+ CD8 T cell subsets correlate with disease protection. Given previous findings suggesting that CD103 and TCF1 expression might be mutually exclusive, we conducted a mIF analysis using nine protein markers and nuclei to confirm tissue residency in a subset of patients from cohort A (n=10). The additional markers included CD69, HOBIT and RUNX3 to confirm tissue residence, and CD45RO as a marker of previous antigenic exposure and memory (figure 2H). We leveraged an artificial intelligence algorithm and histocytometry for quantification and visualization of mIF data (online supplemental file 2). We observed that nearly 100% of CD103+ cells were positive for the additional tissue residency and memory markers, irrespective of TCF1 expression status (figure 2I). This comprehensive analysis provided compelling evidence that the CD103+ CD8 T cells associated with disease control indeed represented TRM cells.

TRM CD8 and cDC1 signatures are associated with survival in the TCGA melanoma database

We explored the clinical relevance of tissue residency evaluating its influence on the survival of melanoma patients. We initially examined the survival outcomes of TCGA melanoma patients (n=472) based on high expression levels of the following markers: CD8, CD103, CD8 and CD103 combined, and the gene signature of cDC1 (CD11c, HLADR, and BDCA3). This analysis revealed an association between enhanced survival and increased expression of these markers, particularly CD8 and cDC1 (online supplemental file 2). We also constructed extended TRM and cDC1 transcriptional signatures (online supplemental file 2). The enrichment of the signatures significantly correlated with improved overall survival probability (TRM Kaplan-Meier p=2.46×10−6; cDC1 p=3.84×10–6) (figure 3A), while the combination of both signatures raised the p value to 3.48X10–7. Notably, the p value of the TRM signature was significantly more robust than that of CD8-high expressors and CD103-high expressors (online supplemental file 2). Altogether, these findings underscore the pivotal role of TRM CD8 T cells in influencing patient outcomes, even when both subsets are compared within patients with high level of infiltration of CD8 T cells (online supplemental file 2).

Figure 3

There is a high correlation between the enrichment of TRM CD8 T cells and cDC1, which is associated with patient survival. (A) Kaplan-Meier survival curves for TRM CD8 T cells (top) and dendritic cell type 1 cDC1 (bottom). Curves show a comparison over the course of 180 months between high versus low enrichment samples. For both populations high enrichment is associated with a better patient survival probability, as demonstrated by the HR and p value obtained from Cox regression. (B) Whole slide images of IF staining for TCF1 (green), CD103 (yellow), CD8 (red), CD11c (magenta), HLA-DR (gray), BDCA-3 (yellow hot), MART-1 (sepia), and nuclei (blue) of representative DF and M patients (top). For the topological maps (middle) representing the X and Y coordinates, all colors were merged as follows: of TCF1+CD103+ CD8 (green), TCF1− CD103+CD8 (blue), cDC1 (orange), and tumor cells (gray). Each point represents one cell. Micrograph (bottom) showing single-color overlap analysis of CD103+CD8+ (red), TCF1 (orange), cDC1 (green), and MART-1 (sepia). In squares of different colors, we show the three way interactions between TCF1− and TCF1+ TRM CD8s, cDC1s, and tumorous cells. To visualize the whole slide images see online supplemental file 2. (C) Bar graphs depicting the percentage of cells localized on intratumorally or stromal areas defined by MART-1 expression. (D) Simple linear regression and Spearman’s correlation test for %TCF1+CD103+ CD8+ vs %cDC1 (left) and %TCF1− CD103+CD8+ vs %cDC1 (right). (E) TRM and cDC1 enrichment association across melanoma patients in the TCGA database is presented by a linear regression analysis.

TRM T cell subsets are in close proximity with cDC1s in tumor and stromal regions

We assessed the spatial relationship between TRM T cells and cDC1s within melanoma and examined whether these immune cells exhibited close proximity suggestive of cell-to-cell interactions. Using mIF staining, we identified TCF1+ and TCF1− TRM CD8 T cells, cDC1s, and melanoma cells. Four whole slides from both DF and M patients were randomly selected and scanned, enabling the generation of topological maps delineating tumor and stromal regions (figure 3B and online supplemental file 2). Our analysis revealed that these immune cell populations were predominantly localized in stromal areas surrounding the tumor. Figure 3B shows an image illustrating single-color representations of TRM T cells, cDC1s, and melanoma cells, demonstrating examples of two-way and three-way cell interactions involving TCF1+ or TCF1− T cells with melanoma cells, as well as TCF1+ or TCF1− T cells with cDC1s and melanoma cells (middle panel). A representative image illustrates the close interactions between TCF1+ TRM CD8, cDC1 and melanoma cells (bottom panel). TCF1− T cells (19.8%) and cDC1s (26.6%) appear to exhibit greater enrichment in tumorous areas compared with TCF1+ T cells (14%) (figure 3C). To explore the relationship between TCF1+ and TCF1− CD103+CD8+ T cells and cDC1s, we conducted a simple linear regression analysis involving all patients from cohort A, observing positive and significant correlations (figure 3D). We also observed a high correlation in patients from the TCGA database (figure 3E). These findings provide compelling evidence for close interactions between both TCF1+ and TCF1− subsets of CD103+ CD8 T cells with cDC1s and tumor cells, suggesting that these three immune populations are controlling the tumor growth.

Figure 4

TCF1+ and TCF1− TRM CD8 T cells exhibit differential functional features. (A) Multiplex immunofluorescence (mIF) staining for CD8 (red), TCF1 (green), CD103 (yellow), PD1 (bright yellow), IFN-γ (magenta), Ki67 (gray), CD45RO (cyan) and nuclei (blue), white square points out to an example of a TCF1+ and yellow square to a TCF1− cell. (B) Histogram (top) showing the mean fluorescence intensity (MFI) of IFN-γ (top) and Ki67 (bottom) of TCF1+ (green) and TCF1− (blue) subsets gated on PD1+ TRM cells. (C, D) Histograms and bar graph showing the MFI and proportion, respectively, of IFN-γ+ and IFN-γ−, and Ki67+ and Ki67− cells in the fraction of PD1+ TRM CD8 T cells that are TCF1+ (C) and TCF1− (D). Patients are divided in disease-free (DF, n=5) and metastatic (M, n=5). (C, D) Thresholds to assign positive and negative cells for IFN-γ and Ki67 are shown in online supplemental file 2. (B–D) Concatenated data of the 5 DF and 5 M datasets. (E) Whole slide images of mIF staining representing PD1+ TRM CD8 cells, marked as TCF1− or TCF1+, apoptotic melanoma cells, phosphorylated caspase 3+ and nuclei of representative DF and M patients. Topological maps are shown in the left panels while mIF images in the right panels. For the left topological map each point represents one cell and populations are marked as GZB+LAMP-1+ PD1+ TRM CD8 cells: TCF1− (blue) and TCF1+ (green), and MART-1+Casp3-p+ apoptotic melanoma cells (sepia). mIF also shows single-color overlap analysis of these cells. Here, GZB+LAMP-1+ PD1+ TRM CD8 cells: TCF1− (are marked as red) and TCF1+ (green), and MART-1+Casp3-p+ apoptotic melanoma cells (sepia), see also online supplemental videos 1 and 2. (F) Distance of TCF1+ and TCF1− PD1+ TRM to apoptotic melanoma cells. Unpaired Student’s t-test, *p<0.05 and **p<0.01.

Melanoma controllers harbor more functionally active TRM T cells

While confirming the TRM nature of CD103+CD8 T cells, we observed differences in the expression of tissue residence-related genes between TCF1-positive and TCF1-negative subsets. The former exhibited varying levels of marker expression, identifying a RUNX3-high population that was absent in their TCF1− counterparts (online supplemental file 2). A similar pattern of expression was observed for HOBIT (online supplemental file 2), indicating intrinsic lower levels of tissue residency markers for the TCF1− CD103+ CD8 T cells.

We conducted a detailed exploration of the differences between TRM subsets, analyzing their effector-function potential and its implications for disease control. We evaluated the expression of the checkpoint inhibitor PD1, the antitumoral cytokine IFN-γ, and the proliferation marker Ki67 (figure 4A and online supplemental file 2). Notably, more than 80% of cells in both TCF1 subsets expressed PD1, suggesting similar states of exhaustion (online supplemental file 2). Initially defined as an unresponsive, terminal state, exhaustion has been redefined through subsequent research, revealing a continuum of states marked by varying levels of effector activities. Gating on the PD1+ TRM subset, we observed that TCF1+ cells not only expressed higher levels of IFN-γ and Ki67, but also harbored a greater number of cells positive for these markers than TCF1− cells, supporting a higher proportion of functionally active cells within this subset (figure 4B and online supplemental file 2). Strikingly, we observed higher levels of both IFN-γ and Ki67 in both TCF1+ and TCF1− TRM subsets in DF patients than in M patients (figure 4C,D). In summary, our results suggest that while the TCF1+ TRM CD8 T cells display enhanced functionality compared with the TCF1− cells, both subsets seem functional since both express effector molecules. While present in low numbers, TRM CD8 T cells in M patients still harbor IFN-γ+ and Ki67+ effector cells.

Cytotoxic TCF1− TRM CD8 T cells associate with apoptotic melanoma cells

After observing the apparent functional differences between TCF1+ and TCF1− TRM subsets, we explored their cytolytic potential and their spatial proximity to melanoma cells. We randomly selected four DF and four M patients for mIF staining of GZB+ (granzyme B) and LAMP-1+ (lysosomal associated membrane protein 1) markers, indicative of cytotoxicity potential within the PD1+TRM subsets. Additionally, Casp-3-p (phosphorylated caspase 3) and MART-1 markers were assessed to determine whether cytotoxic cells were spatially associated with apoptotic melanoma cells. We observed that GZB+LAMP-1+ PD1+ cells were enriched in the TCF1− TRM subset, with DF patients showing a higher proportion (64%) compared with M patients (47%) (online supplemental file 2). Since this analysis considered the entire field without regard to tumor location, we generated XY coordinate plots to assess spatial associations between cytotoxic TRM subsets and apoptotic melanoma cells. The TCF1− subset was enriched in tumor areas, exhibiting closer proximity to Casp3-p+MART-1+ cells (figure 4E and online supplemental file 2). Image processing facilitated the visualization of interactions between cytotoxic cells and apoptotic melanoma cells (figure 4E and supplemental videos 1-2). Distance estimation between Casp3-p+MART-1+ cells and cytotoxic TRM subsets revealed that the TCF1− population was significantly closer to apoptotic tumor cells, supporting a more direct involvement in tumor cell death (figure 4F).

TCF1− TRM T cells accumulate exhaustion markers in metastatic patients

Exhausted populations have been recognized to transition from a progenitor-like population expressing TCF1 to a more terminally exhausted population, characterized by negativity for TCF1 and positivity for inhibitory checkpoint receptors PD1, TIM3, LAG3, CTLA4, and CD39. Given the prevalent expression of PD1 in TRM CD8 T cells, we explored whether the TCF1-positive and TCF1-negative TRM subsets might exhibit signs of terminal exhaustion. To investigate this, we assessed the expression of PD1, TIM3, the exhaustion-associated transcription factor TOX, and CXCR5, a marker of functionality among exhausted cells and of positive response to CBI (figure 5A and online supplemental file 2). Our analysis revealed that both TCF1+ and TCF1− TRM subsets contained exhausted cells coexpressing PD1 and TOX (figure 5B). The TCF1+ population displayed higher levels of CXCR5 expression and a greater proportion of CXCR5-positive cells, aligning more closely with a progenitor-like effector state while the TCF1− popul